International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 5

ISSN: 2321-8169
1255 1259

______________________________________________________________________________________________

Advance Method for Brain Tumor Classification

Dipanshu N. Masalkar,
Research Student

Mr. Shitole A.S.,

Assistant Professor

Dept. of Computer Engineering

Sinhgad Academy of Eneinrreing, Kondhwa, Pune, India
dipanshuchoude@gmail.com

Dept. of computer Engineering

Sinhgad Academy of Eneinrreing, Kondhwa, Pune, India
shitole_ajit@yahoo.com

Abstract Detection and classification of tumor from MRI brain image is becoming most challenging area to research. This paper describe the
proposed system for brain tumor detection and classification along with the help of Artificial neural network. This proposed system includes
several steps segmentation for tumor detection , feature extraction , dimensionality reduction of extracted feature for removing redundant
features and classification of tumor. Previously this decision is taken manually by humans with the help of MR (Magnetic resonance) or CT
(Computerized Tomography) scan image of brain. But the time require for this is more and result may not be accurate. Image may contain some
noise due to error in machine performance which will result in inaccuracy and becomes hazardous to patient suffering from this disease.
Artificial intelligence and image processing have enormous growth in medical research field with the help of neural network and fuzzy logic. In
this paper Probabilistic Neural Network is used for decision making which is followed by Image Preprocessing with Gaussian Filter Method,
Image segmentation and detection by using Threshold Based Segmentation Method, Feature extraction by Gray Level Co-occurrence Matrix
(GLCM)features, and Dimensionality reduction by Principal Component Analysis. Main advantage of this method is that it will give fast and
accurate result with the help of training data set and it reduces time and computation power.

Keywords- Brain tumor classification, Probabilistic Neural Network, Gaussian Filter, Gray Threshold Based Segmentation, Gray Level Cooccurrence Matrix, Principal Component Analysis.

__________________________________________________*****_________________________________________________
I.

INTRODUCTION

Abnormal growth of cells developed inside human body is

called as Tumor. Brain Tumor is an intra-cranial solid neoplasm
occurs within the brain or the central spinal canal. Brain tumor
is implicitly serious and life-ominous disease because brain is
very fragile part of human body to treat for. However, Brain
tumors can be malignant that is cancerous or benign that is
non-cancerous. Treatment of brain tumor depends on proper
diagnosis and depends on the different factor like the type of
tumor, location, size and state of development.
MRI is technique used to measuring density of photons in
tissue; it is based on fundamental property of photon that spins
and possesses magnetic movement. It is done to visualize the
internal structure of human body, gives superior image quality.
Useful information is reveled from MR image and it helps in
proper diagnosis. MR image of brain gives only the visual
internal picture of the brain. Fig 1 shows the normal MR brain
image and image with tumor. Early and proper detection of
tumor is the key for the proper treatment. Previously stage of
tumor is used to be detected manually with the help of
observation of image by doctors and sometimes it takes more
time and sometimes results may inaccurate. There are many
different types of brain tumor and only experienced and expert
doctor can able to give the accurate result. So we require
accurate diagnosis tool for proper treatment. Proper diagnosis
tool should include detection, segmentation and classification
of MRI brain image. Detection involves finding the presence
of tumor; segmentation involves the detection of size and

location of tumor and classification involves the detection of

stage of tumor.

Now a days many computer added tool is used in medical

field. These tools possess a property of quick and accurate
result. Using the benefit of Artificial intelligence and image
processing a tool that will detect and classify the brain tumor
can be developed. Input image to this system is Brain MR
Image which is taken by MRI scanner.
Remaining paper is organized as organized as follows II gives
Literature Survey,III proposed method, IV explains Algorithm
for tumor classification, V Computational Details And Expected
Result and ends with VI Conclusion.
II.

LITERATURE SURVEY

In various recent research on segmentation , feature extraction

and classification of MR brain image many different
combination of segmentation and classification is used by
various auther.
1255

IJRITCC | May 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 5

ISSN: 2321-8169
1255 1259

______________________________________________________________________________________________
Discrete cosine transform for segmentation and Probabilistic
neural network for classification is developed in matlab by
author D. Sridhar and evaluation was performed on 20 set of
images dividing in to different combination of test set and
training set of data.
Where as PCA for feature extraction is used along with the
PNN by author Mohd Fauzi Othman and evaluation was
performed on 20 set of images .In system proposed by author
C. Nageswara rao NN and K-NN classifier is used for
classification. In this double classification is used for accuracy
check . where accuracy given by NN is from 66% to 98.2 %
and by K-NN 80% to 100%. Where in some system SVM is
used for classification.
Reference [4] and [5] explained the different ANN algorithm
for the brain image classification.
III.

Test set of MR Brain

image

Preprocessing By
Gaussian Filter

Preprocessing By
Gaussian Filter

Threshold Based
segmentation

GLCM Feature Extraction

PROPOSED METHOD

In this proposed system the work carried out in various stages.

System starts with the input of Training set of MR brain image
data of different types of tumor and test set of MR brain image
data of normal brain or abnormal brain. In next stage
preprocessing is done on both images to remove noise. Then
feature extraction is done for training set of data, then
dimensionality reduced to remove redundant feature and saved
in hidden layer of PNN for classification. Segmentation is
done ontest set of data to detect presence of tumor in test set of
data. If present then it goes for next step that is feature
extraction and dimensionality reduction of feature to remove
redundant feature. Final result will give classification of test
set of data. Figure 2 shows proposed system architecture.
IV.

Train set of MR Brain

image

Dimensionality Reduction By
PCA

PNN Classifier

Result

Figure 2: System Architecture

ALGORITHMS FOR TUMOR CLASSIFICATION

A. Gaussian Filter
Sometimes images captured are corrupted and contains
noise due to random variations in intensity, illumination, or
have poor contrast and cant be used directly. So to remove
the noise from MR image Gaussian Filter is used. Therefore
transformation of pixel intensity values to obtain certain
image characteristics, enhancement for improving contrast and
Smoothing to remove noises as well as template matching to
detectsknown patterns can be done by using Gaussian filter
method. In general sliding mean or box filter noise reduction
method is used in which the pixel value is replaced by the
mean of neighbouring pixel values. Gaussian Filter method is
similar to this but neighbouring pixel values is given with
different weighting defined by spatial Gaussian distribution.
In this method the Gaussian function at every point of image is
non zero which means that each pixel of image is calculated
using this method.
Formula for calculating value of pixel value for 2-D Gaussian
Filter is as fallows [13].

Formula for calculating value of pixel value for 1-D Gaussian

Filter is as fallows [13].

Where x is distance from the origin in the horizontal axis and

y isdistance from the origin in the verticalaxis. is standard
deviation of Gaussian Distribution.
B. Threshold Based segmentation
Presence of tumor is detected by Threshold based
segmentation method. Segmentation is done by using pixel
based Thresholding method. In this method intensity value of
pixel is used for partition. In gray scale image Threshold value
is given to threshold function and value below is change to 1
for representing tumor and 0 for representing the background.
Formula for representing threshold value is given as

Where v represents gray value and t represents threshold

value.
More than one object can be shown in segmentation using
Multiple Threshold Based Segmentation Method. In which
1256
IJRITCC | May 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 5

ISSN: 2321-8169
1255 1259

______________________________________________________________________________________________
multiple threshold values are used to represent pixel intensity.
And it is given by

C. Gray Level Co-occurrence MatrixFigures and Tables

Gray Level Co-occurrence Matrix (GLCM) is used for
feature extraction of MR image. Feature of image based on
pixels and its neighboring pixels are extracted by image
GLCM matrix is formed contains the textural feature based on
two pixel intensity values. Details about overall texture
content of image are given by parameters calculated by the
GLCM of original gray image. Feature based on pixel and its
neighboring pixel is extracted by GLCM (i, j). It is a two
dimensional function ,composed of n horizontal direction
pixels and m of vertical direction pixels and horizontal and
vertical co -ordinates of the image is given by i , j. 0 i m
j m where total pixel number is m n. First the intensity of
pixel and its neighbouring pixel is calculated for entire image.
For getting more reliable texture feature multiple GLCMs are
computed for different Rotations of at 0, 45, 90, and 135
which can give spatial relationship between neighbouring
pixel. After calculation of GLCMs of image, it is used to
calculate contrast, energy, homogeneity, and co-relation of
image which uniquely describes the images. This method
reduces the computational complexity.
Formula for calculating contrast intensity is[3]
Contrast=

Energy is measured by[3]

Energy=
Homogeneity gives the spatial distribution and closeness of
gray level[3]
Homogeneity=
Correlation is gives the relation between the pixel and its
neighbouring pixel according to its contrast, energy and
homogeneity [3].

component (PC). PC is the observed variable which is

optimally weighted and linearly combined. It is the linear
transform which rotate the axis of image space along line of
maximum variance. Greatest amount of variation variable
combination give the first PCA. Second PCA give next
greatest variable combination and which is derived from the
first PCA. And there can be many possible rotations for
reduction.
E. Probabilistic Neural Network
In this stage the test MR image is compared with the training
MR image and gives output training MR image which is
similar to test image. For classification purpose Probabilistic
Neural Network (PNN) is used. PNN is supervised feed
forward neural network derived from Bayes classifiers. Where
Parzen calculator is used for classification which uses
probability density function (pdf) for each class of training
sample. If training sample is increased then classification goes
near to density function of the class.
For each class using training sample on the basis of pdf
Parzen calculator takes a classification. The Bayes rule for
classification for class i is given by
Where
and
is the prior probability of presence of
identical feature of class i and j. And
and
is pdf of
class i and j.
Where
is the submission of Gaussian probability
distribution function.
. Architecture of PNN consists of four layer input layer,
pattern layer, submission layer, Output layer.
1. Input layer
Feature vector of training dataset and test data set is given
to the input layer. There are many nodes to the input layer
which is connected to the pattern layer.
2. Pattern layer
To each training input layer node pattern layer node is
connected. Pattern node consists of processing element and it
should be trained once. Count of processing element should be
same for better classification. Pattern node is connected to
each class of weight vector which is multiplied with the input
vector. Weight vector is trained vector. This will do a
comparison between input vector and processing element.
Product results in to the closeness of the class. Following is
the equation for calculating product. Where
is weight of
the processing element and
is input feature from input layer
connected to that particular pattern node.

Correlation=
Where
is the mean and
of GLCM x row and y column.

is the standard deviation

D. Principal Component Analysis

In this step we are going to reduce the dimension of MRI
image. Images though small in size are having large
dimensionality this leads to very large computational time,
complexity and memory occupation. This is done by using
Principle component Analysis Method. In this step feature
extracted from GLCM is given to PCA. Data that is redundant
can be reducing to artificial variables called as principal

This
should be passed through the activation node
calculated by formula given below.
3. Submission layer
Output from pattern is given to the submission node. It
simply gives the submissions of activated class pattern. And it
is given by following formula.

1257
IJRITCC | May 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 5

ISSN: 2321-8169
1255 1259

______________________________________________________________________________________________
Where N is number of training set example is hidden node
activation is smoothing factor.
Feature vector of training and test set of data will be the
input to the PNN classifier, following are steps of PNN
algorithm
Feature vector of training set is available and the class to
which it belongs is defined.
The following are the steps for setting up the PNN.
1. Read in the feature vector and the defined class number.
2. Sort them into the x sets where each set should contain
single class of vectors.
3. Gaussian function to be defined for each x set which will
concentrate on each feature vector in set x. Define the summed
Gaussian output function.
Once the PNN is set, then feature vector from test data set is
given to it for classification as follows.
1. Gaussian function in each class to be fed by input feature
vector.
2. Gaussian functional values are computed at the hidden
nodes from each set of hidden node.
3. Gaussian functional values of hidden set of node is fed to
the single outputnode for that group
4. At each class output node, submission all of the inputs is
done and multiply by constant
5. Output node will find maximum value of all functional
values that is summed.
V.

COMPUTATIONAL DETAILS AND EXPECTED RESULT

Figure 3 : PNN Architecture

In this process testing will be performed on different number

of sample training data set and sample test data set.
Classfication technique PNN will be applied to generate
resultant output as shown in fig 4.
Test Image

Equivalent Image

Let N be the number of training set of images

N={N1, N2, ., Nn}.
T be the test set of images
T={T1,T2,..Tn}
After getting feature vector of N And T images , Feature
vecture of training set of data is classified in to K Set of
classification at PNN classification stage.
K={K1, K2, , Kn}.
Each class contain different feature vector F.
FK1={F11, F12,., F1n}
FK2={F21, F22, . , F2n}
.
.
FKn={Fn1, Fn2, . , Fnn}.

equivalent image of Class 1

And depending on the feature vector vector of test image

which will be compaired with the feature vector of training
image output of classification is given.

equivalent image of Class 2

equivalent image of Class 3

Fig. 4 Expected Resultant Output

1)
1258
IJRITCC | May 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________

International Journal on Recent and Innovation Trends in Computing and Communication

Volume: 2 Issue: 5

ISSN: 2321-8169
1255 1259

______________________________________________________________________________________________
VI.

CONCLUSION

In this paper method for brain tumor detection and

classification is proposed with the help of Gaussian Filter,
Threshold Based Segmentation, Gray Level Co-occurrence Matrix
(GLCM)features, Principal Component Analysis and Probabilistic

Neural Network. This proposed method will detect as well as

classify the MR image. If test image is not similar to any
training image then image is included in training set data. This
method will give fast and more accurate result and helps
physician for proper treatment.

[7]

[8]

[9]

[10]
ACKNOWLEDGEMENT
I would like to express my deep gratitude to Mr. Shitole A. S.
my research guide, for his patient guidance enthusiastic
encouragement and useful critiques for this research work. I
would also like to thank Prof. B. B. Gite(Head of Department)
and Prof. L.J.Sankpal(M.E Co-ordinator) for their advice and
assistance in keeping my progress on schedule. Finally, I wish
to thank my parents for their support and encouragement
throughout my study.

[1]

[2]

[3]

[4]

[5]

[6]

REFERENCES
D. SRIDHAR and N. MURALI KRISHNA Brain
Tumor Classification U sing Discrete Cosine
Transform and Probabilistic Neural Network 2013
International Conference on Signal Processing, Image
Processing and Pattern Recognition,978-1-46734862-1/13/$31.00 2013 IEEE
Mohd Fauzi Othman and Mohd Ariffanan Mohd
Basri Probabilistic Neural NetworkFor Brain Tumor
Classification , 2011 Second International
Conference on Intelligent Systems, Modelling and
Simulation, 978-0-7695-4336-9/11 $26.00 2011
IEEE.
C.Nageswara Rao, S.Sreehari Sastry,K.Mallika, Ha
Sie Tiong and K.B.MahalakshmiCo-Occurrence
Matrix and Its Statistical Features as an Approach for
Identification Of Phase Transitions Of Mesogens,
International Journal of Innovative Research in
Science, Engineering and Technology, ISSN: 23198753
Amer Al-Badarneh, Hassan Najadat, Ali M. Alraziqi
A Classifier to Detect Tumor Disease in MRI Brain
Images, 2012 IEEE/ACM International Conference
on Advances in Social Networks Analysis and
Mining, 978-0-7695-4799-2/12 $26.00 2012 IEEE.
P.
Jeatrakul
and
K.W.
Wong
ComparingthePerformanceofDifferentNeuralNetwor
ksforBinaryClassificationProblems,
2009Eighth
International Symposium onNatural Language
Processing
J.
Papitha,
G.
Merlin
Nancyand
D.
Nedumaran,Member,
IEEE
Compression
Techniques onMR ImageA ComparativeStudy,
International conference on Communicationand
Signal Processing, April 3-5, 2013, India.

[11]

[12]

[13]

Madhuri A. Joshi, Digital image Processing An

Algorithmic Approach, Book PHI Learning and
Private Limited.
Anil K. Joshi Fundamentals of Digital image
processing, Book-Prentice Hall of India Priveted
Limited.
Pv Nageswara T Uma Devi1, Dsvgk Kaladhar1,Gr
Sridhar2, Allam Appa Rao3, A Probabilistic Neural
Network
Approach
ForProtein
Superfamily
Classification, Journal of Theoretical and Applied
Information Technology, 2005 - 2009 JATIT. All
rights reserved.
K.S.Deepak, K.Gokul, R.Hinduja, S.Rajkumar, An
Efficient Approach to Predict Tumor in 2d
BrainImage Using Classification Techniques.
S.N.Deepa & B.Aruna Devi,Artificial Neural
Networks design for Classification ofBrain Tumor,
2012 International Conference on Computer
Communication and Informatics (ICCCI -2012), Jan.
10 12, 2012, Coimbatore, INDIA
Prof.
Vikas
Gupta,
Kaustubh
S.
Sagale,Implementation of Classification System for
Brain Cancer Using Back propagation Network And
Mri, 2012 Nirma University International
Conference On Engineering, Nuicone-2012, 0608december, 2012.
G. Deng and L. W. CahillAn Adaptive Gaussian
Filter For Noise Reduction and Edge Detection,
Department of Electronic Engineering, La Trobe
University Bundoora Victoria 3083 Australia

1259
IJRITCC | May 2014, Available @ http://www.ijritcc.org

_______________________________________________________________________________________